xref: /linux/arch/x86/kvm/i8254.c (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
1 /*
2  * 8253/8254 interval timer emulation
3  *
4  * Copyright (c) 2003-2004 Fabrice Bellard
5  * Copyright (c) 2006 Intel Corporation
6  * Copyright (c) 2007 Keir Fraser, XenSource Inc
7  * Copyright (c) 2008 Intel Corporation
8  * Copyright 2009 Red Hat, Inc. and/or its affiliates.
9  *
10  * Permission is hereby granted, free of charge, to any person obtaining a copy
11  * of this software and associated documentation files (the "Software"), to deal
12  * in the Software without restriction, including without limitation the rights
13  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
14  * copies of the Software, and to permit persons to whom the Software is
15  * furnished to do so, subject to the following conditions:
16  *
17  * The above copyright notice and this permission notice shall be included in
18  * all copies or substantial portions of the Software.
19  *
20  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
21  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
23  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
24  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
25  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26  * THE SOFTWARE.
27  *
28  * Authors:
29  *   Sheng Yang <sheng.yang@intel.com>
30  *   Based on QEMU and Xen.
31  */
32 
33 #define pr_fmt(fmt) "pit: " fmt
34 
35 #include <linux/kvm_host.h>
36 #include <linux/slab.h>
37 
38 #include "irq.h"
39 #include "i8254.h"
40 #include "x86.h"
41 
42 #ifndef CONFIG_X86_64
43 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
44 #else
45 #define mod_64(x, y) ((x) % (y))
46 #endif
47 
48 #define RW_STATE_LSB 1
49 #define RW_STATE_MSB 2
50 #define RW_STATE_WORD0 3
51 #define RW_STATE_WORD1 4
52 
53 /* Compute with 96 bit intermediate result: (a*b)/c */
54 static u64 muldiv64(u64 a, u32 b, u32 c)
55 {
56 	union {
57 		u64 ll;
58 		struct {
59 			u32 low, high;
60 		} l;
61 	} u, res;
62 	u64 rl, rh;
63 
64 	u.ll = a;
65 	rl = (u64)u.l.low * (u64)b;
66 	rh = (u64)u.l.high * (u64)b;
67 	rh += (rl >> 32);
68 	res.l.high = div64_u64(rh, c);
69 	res.l.low = div64_u64(((mod_64(rh, c) << 32) + (rl & 0xffffffff)), c);
70 	return res.ll;
71 }
72 
73 static void pit_set_gate(struct kvm *kvm, int channel, u32 val)
74 {
75 	struct kvm_kpit_channel_state *c =
76 		&kvm->arch.vpit->pit_state.channels[channel];
77 
78 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
79 
80 	switch (c->mode) {
81 	default:
82 	case 0:
83 	case 4:
84 		/* XXX: just disable/enable counting */
85 		break;
86 	case 1:
87 	case 2:
88 	case 3:
89 	case 5:
90 		/* Restart counting on rising edge. */
91 		if (c->gate < val)
92 			c->count_load_time = ktime_get();
93 		break;
94 	}
95 
96 	c->gate = val;
97 }
98 
99 static int pit_get_gate(struct kvm *kvm, int channel)
100 {
101 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
102 
103 	return kvm->arch.vpit->pit_state.channels[channel].gate;
104 }
105 
106 static s64 __kpit_elapsed(struct kvm *kvm)
107 {
108 	s64 elapsed;
109 	ktime_t remaining;
110 	struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
111 
112 	if (!ps->period)
113 		return 0;
114 
115 	/*
116 	 * The Counter does not stop when it reaches zero. In
117 	 * Modes 0, 1, 4, and 5 the Counter ``wraps around'' to
118 	 * the highest count, either FFFF hex for binary counting
119 	 * or 9999 for BCD counting, and continues counting.
120 	 * Modes 2 and 3 are periodic; the Counter reloads
121 	 * itself with the initial count and continues counting
122 	 * from there.
123 	 */
124 	remaining = hrtimer_get_remaining(&ps->timer);
125 	elapsed = ps->period - ktime_to_ns(remaining);
126 
127 	return elapsed;
128 }
129 
130 static s64 kpit_elapsed(struct kvm *kvm, struct kvm_kpit_channel_state *c,
131 			int channel)
132 {
133 	if (channel == 0)
134 		return __kpit_elapsed(kvm);
135 
136 	return ktime_to_ns(ktime_sub(ktime_get(), c->count_load_time));
137 }
138 
139 static int pit_get_count(struct kvm *kvm, int channel)
140 {
141 	struct kvm_kpit_channel_state *c =
142 		&kvm->arch.vpit->pit_state.channels[channel];
143 	s64 d, t;
144 	int counter;
145 
146 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
147 
148 	t = kpit_elapsed(kvm, c, channel);
149 	d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
150 
151 	switch (c->mode) {
152 	case 0:
153 	case 1:
154 	case 4:
155 	case 5:
156 		counter = (c->count - d) & 0xffff;
157 		break;
158 	case 3:
159 		/* XXX: may be incorrect for odd counts */
160 		counter = c->count - (mod_64((2 * d), c->count));
161 		break;
162 	default:
163 		counter = c->count - mod_64(d, c->count);
164 		break;
165 	}
166 	return counter;
167 }
168 
169 static int pit_get_out(struct kvm *kvm, int channel)
170 {
171 	struct kvm_kpit_channel_state *c =
172 		&kvm->arch.vpit->pit_state.channels[channel];
173 	s64 d, t;
174 	int out;
175 
176 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
177 
178 	t = kpit_elapsed(kvm, c, channel);
179 	d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
180 
181 	switch (c->mode) {
182 	default:
183 	case 0:
184 		out = (d >= c->count);
185 		break;
186 	case 1:
187 		out = (d < c->count);
188 		break;
189 	case 2:
190 		out = ((mod_64(d, c->count) == 0) && (d != 0));
191 		break;
192 	case 3:
193 		out = (mod_64(d, c->count) < ((c->count + 1) >> 1));
194 		break;
195 	case 4:
196 	case 5:
197 		out = (d == c->count);
198 		break;
199 	}
200 
201 	return out;
202 }
203 
204 static void pit_latch_count(struct kvm *kvm, int channel)
205 {
206 	struct kvm_kpit_channel_state *c =
207 		&kvm->arch.vpit->pit_state.channels[channel];
208 
209 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
210 
211 	if (!c->count_latched) {
212 		c->latched_count = pit_get_count(kvm, channel);
213 		c->count_latched = c->rw_mode;
214 	}
215 }
216 
217 static void pit_latch_status(struct kvm *kvm, int channel)
218 {
219 	struct kvm_kpit_channel_state *c =
220 		&kvm->arch.vpit->pit_state.channels[channel];
221 
222 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
223 
224 	if (!c->status_latched) {
225 		/* TODO: Return NULL COUNT (bit 6). */
226 		c->status = ((pit_get_out(kvm, channel) << 7) |
227 				(c->rw_mode << 4) |
228 				(c->mode << 1) |
229 				c->bcd);
230 		c->status_latched = 1;
231 	}
232 }
233 
234 static void kvm_pit_ack_irq(struct kvm_irq_ack_notifier *kian)
235 {
236 	struct kvm_kpit_state *ps = container_of(kian, struct kvm_kpit_state,
237 						 irq_ack_notifier);
238 	int value;
239 
240 	spin_lock(&ps->inject_lock);
241 	value = atomic_dec_return(&ps->pending);
242 	if (value < 0)
243 		/* spurious acks can be generated if, for example, the
244 		 * PIC is being reset.  Handle it gracefully here
245 		 */
246 		atomic_inc(&ps->pending);
247 	else if (value > 0)
248 		/* in this case, we had multiple outstanding pit interrupts
249 		 * that we needed to inject.  Reinject
250 		 */
251 		queue_kthread_work(&ps->pit->worker, &ps->pit->expired);
252 	ps->irq_ack = 1;
253 	spin_unlock(&ps->inject_lock);
254 }
255 
256 void __kvm_migrate_pit_timer(struct kvm_vcpu *vcpu)
257 {
258 	struct kvm_pit *pit = vcpu->kvm->arch.vpit;
259 	struct hrtimer *timer;
260 
261 	if (!kvm_vcpu_is_bsp(vcpu) || !pit)
262 		return;
263 
264 	timer = &pit->pit_state.timer;
265 	if (hrtimer_cancel(timer))
266 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
267 }
268 
269 static void destroy_pit_timer(struct kvm_pit *pit)
270 {
271 	hrtimer_cancel(&pit->pit_state.timer);
272 	flush_kthread_work(&pit->expired);
273 }
274 
275 static void pit_do_work(struct kthread_work *work)
276 {
277 	struct kvm_pit *pit = container_of(work, struct kvm_pit, expired);
278 	struct kvm *kvm = pit->kvm;
279 	struct kvm_vcpu *vcpu;
280 	int i;
281 	struct kvm_kpit_state *ps = &pit->pit_state;
282 	int inject = 0;
283 
284 	/* Try to inject pending interrupts when
285 	 * last one has been acked.
286 	 */
287 	spin_lock(&ps->inject_lock);
288 	if (ps->irq_ack) {
289 		ps->irq_ack = 0;
290 		inject = 1;
291 	}
292 	spin_unlock(&ps->inject_lock);
293 	if (inject) {
294 		kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 1, false);
295 		kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 0, false);
296 
297 		/*
298 		 * Provides NMI watchdog support via Virtual Wire mode.
299 		 * The route is: PIT -> PIC -> LVT0 in NMI mode.
300 		 *
301 		 * Note: Our Virtual Wire implementation is simplified, only
302 		 * propagating PIT interrupts to all VCPUs when they have set
303 		 * LVT0 to NMI delivery. Other PIC interrupts are just sent to
304 		 * VCPU0, and only if its LVT0 is in EXTINT mode.
305 		 */
306 		if (kvm->arch.vapics_in_nmi_mode > 0)
307 			kvm_for_each_vcpu(i, vcpu, kvm)
308 				kvm_apic_nmi_wd_deliver(vcpu);
309 	}
310 }
311 
312 static enum hrtimer_restart pit_timer_fn(struct hrtimer *data)
313 {
314 	struct kvm_kpit_state *ps = container_of(data, struct kvm_kpit_state, timer);
315 	struct kvm_pit *pt = ps->kvm->arch.vpit;
316 
317 	if (ps->reinject || !atomic_read(&ps->pending)) {
318 		atomic_inc(&ps->pending);
319 		queue_kthread_work(&pt->worker, &pt->expired);
320 	}
321 
322 	if (ps->is_periodic) {
323 		hrtimer_add_expires_ns(&ps->timer, ps->period);
324 		return HRTIMER_RESTART;
325 	} else
326 		return HRTIMER_NORESTART;
327 }
328 
329 static void create_pit_timer(struct kvm *kvm, u32 val, int is_period)
330 {
331 	struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
332 	s64 interval;
333 
334 	if (!irqchip_in_kernel(kvm) || ps->flags & KVM_PIT_FLAGS_HPET_LEGACY)
335 		return;
336 
337 	interval = muldiv64(val, NSEC_PER_SEC, KVM_PIT_FREQ);
338 
339 	pr_debug("create pit timer, interval is %llu nsec\n", interval);
340 
341 	/* TODO The new value only affected after the retriggered */
342 	hrtimer_cancel(&ps->timer);
343 	flush_kthread_work(&ps->pit->expired);
344 	ps->period = interval;
345 	ps->is_periodic = is_period;
346 
347 	ps->timer.function = pit_timer_fn;
348 	ps->kvm = ps->pit->kvm;
349 
350 	atomic_set(&ps->pending, 0);
351 	ps->irq_ack = 1;
352 
353 	/*
354 	 * Do not allow the guest to program periodic timers with small
355 	 * interval, since the hrtimers are not throttled by the host
356 	 * scheduler.
357 	 */
358 	if (ps->is_periodic) {
359 		s64 min_period = min_timer_period_us * 1000LL;
360 
361 		if (ps->period < min_period) {
362 			pr_info_ratelimited(
363 			    "kvm: requested %lld ns "
364 			    "i8254 timer period limited to %lld ns\n",
365 			    ps->period, min_period);
366 			ps->period = min_period;
367 		}
368 	}
369 
370 	hrtimer_start(&ps->timer, ktime_add_ns(ktime_get(), interval),
371 		      HRTIMER_MODE_ABS);
372 }
373 
374 static void pit_load_count(struct kvm *kvm, int channel, u32 val)
375 {
376 	struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
377 
378 	WARN_ON(!mutex_is_locked(&ps->lock));
379 
380 	pr_debug("load_count val is %d, channel is %d\n", val, channel);
381 
382 	/*
383 	 * The largest possible initial count is 0; this is equivalent
384 	 * to 216 for binary counting and 104 for BCD counting.
385 	 */
386 	if (val == 0)
387 		val = 0x10000;
388 
389 	ps->channels[channel].count = val;
390 
391 	if (channel != 0) {
392 		ps->channels[channel].count_load_time = ktime_get();
393 		return;
394 	}
395 
396 	/* Two types of timer
397 	 * mode 1 is one shot, mode 2 is period, otherwise del timer */
398 	switch (ps->channels[0].mode) {
399 	case 0:
400 	case 1:
401         /* FIXME: enhance mode 4 precision */
402 	case 4:
403 		create_pit_timer(kvm, val, 0);
404 		break;
405 	case 2:
406 	case 3:
407 		create_pit_timer(kvm, val, 1);
408 		break;
409 	default:
410 		destroy_pit_timer(kvm->arch.vpit);
411 	}
412 }
413 
414 void kvm_pit_load_count(struct kvm *kvm, int channel, u32 val, int hpet_legacy_start)
415 {
416 	u8 saved_mode;
417 	if (hpet_legacy_start) {
418 		/* save existing mode for later reenablement */
419 		saved_mode = kvm->arch.vpit->pit_state.channels[0].mode;
420 		kvm->arch.vpit->pit_state.channels[0].mode = 0xff; /* disable timer */
421 		pit_load_count(kvm, channel, val);
422 		kvm->arch.vpit->pit_state.channels[0].mode = saved_mode;
423 	} else {
424 		pit_load_count(kvm, channel, val);
425 	}
426 }
427 
428 static inline struct kvm_pit *dev_to_pit(struct kvm_io_device *dev)
429 {
430 	return container_of(dev, struct kvm_pit, dev);
431 }
432 
433 static inline struct kvm_pit *speaker_to_pit(struct kvm_io_device *dev)
434 {
435 	return container_of(dev, struct kvm_pit, speaker_dev);
436 }
437 
438 static inline int pit_in_range(gpa_t addr)
439 {
440 	return ((addr >= KVM_PIT_BASE_ADDRESS) &&
441 		(addr < KVM_PIT_BASE_ADDRESS + KVM_PIT_MEM_LENGTH));
442 }
443 
444 static int pit_ioport_write(struct kvm_io_device *this,
445 			    gpa_t addr, int len, const void *data)
446 {
447 	struct kvm_pit *pit = dev_to_pit(this);
448 	struct kvm_kpit_state *pit_state = &pit->pit_state;
449 	struct kvm *kvm = pit->kvm;
450 	int channel, access;
451 	struct kvm_kpit_channel_state *s;
452 	u32 val = *(u32 *) data;
453 	if (!pit_in_range(addr))
454 		return -EOPNOTSUPP;
455 
456 	val  &= 0xff;
457 	addr &= KVM_PIT_CHANNEL_MASK;
458 
459 	mutex_lock(&pit_state->lock);
460 
461 	if (val != 0)
462 		pr_debug("write addr is 0x%x, len is %d, val is 0x%x\n",
463 			 (unsigned int)addr, len, val);
464 
465 	if (addr == 3) {
466 		channel = val >> 6;
467 		if (channel == 3) {
468 			/* Read-Back Command. */
469 			for (channel = 0; channel < 3; channel++) {
470 				s = &pit_state->channels[channel];
471 				if (val & (2 << channel)) {
472 					if (!(val & 0x20))
473 						pit_latch_count(kvm, channel);
474 					if (!(val & 0x10))
475 						pit_latch_status(kvm, channel);
476 				}
477 			}
478 		} else {
479 			/* Select Counter <channel>. */
480 			s = &pit_state->channels[channel];
481 			access = (val >> 4) & KVM_PIT_CHANNEL_MASK;
482 			if (access == 0) {
483 				pit_latch_count(kvm, channel);
484 			} else {
485 				s->rw_mode = access;
486 				s->read_state = access;
487 				s->write_state = access;
488 				s->mode = (val >> 1) & 7;
489 				if (s->mode > 5)
490 					s->mode -= 4;
491 				s->bcd = val & 1;
492 			}
493 		}
494 	} else {
495 		/* Write Count. */
496 		s = &pit_state->channels[addr];
497 		switch (s->write_state) {
498 		default:
499 		case RW_STATE_LSB:
500 			pit_load_count(kvm, addr, val);
501 			break;
502 		case RW_STATE_MSB:
503 			pit_load_count(kvm, addr, val << 8);
504 			break;
505 		case RW_STATE_WORD0:
506 			s->write_latch = val;
507 			s->write_state = RW_STATE_WORD1;
508 			break;
509 		case RW_STATE_WORD1:
510 			pit_load_count(kvm, addr, s->write_latch | (val << 8));
511 			s->write_state = RW_STATE_WORD0;
512 			break;
513 		}
514 	}
515 
516 	mutex_unlock(&pit_state->lock);
517 	return 0;
518 }
519 
520 static int pit_ioport_read(struct kvm_io_device *this,
521 			   gpa_t addr, int len, void *data)
522 {
523 	struct kvm_pit *pit = dev_to_pit(this);
524 	struct kvm_kpit_state *pit_state = &pit->pit_state;
525 	struct kvm *kvm = pit->kvm;
526 	int ret, count;
527 	struct kvm_kpit_channel_state *s;
528 	if (!pit_in_range(addr))
529 		return -EOPNOTSUPP;
530 
531 	addr &= KVM_PIT_CHANNEL_MASK;
532 	if (addr == 3)
533 		return 0;
534 
535 	s = &pit_state->channels[addr];
536 
537 	mutex_lock(&pit_state->lock);
538 
539 	if (s->status_latched) {
540 		s->status_latched = 0;
541 		ret = s->status;
542 	} else if (s->count_latched) {
543 		switch (s->count_latched) {
544 		default:
545 		case RW_STATE_LSB:
546 			ret = s->latched_count & 0xff;
547 			s->count_latched = 0;
548 			break;
549 		case RW_STATE_MSB:
550 			ret = s->latched_count >> 8;
551 			s->count_latched = 0;
552 			break;
553 		case RW_STATE_WORD0:
554 			ret = s->latched_count & 0xff;
555 			s->count_latched = RW_STATE_MSB;
556 			break;
557 		}
558 	} else {
559 		switch (s->read_state) {
560 		default:
561 		case RW_STATE_LSB:
562 			count = pit_get_count(kvm, addr);
563 			ret = count & 0xff;
564 			break;
565 		case RW_STATE_MSB:
566 			count = pit_get_count(kvm, addr);
567 			ret = (count >> 8) & 0xff;
568 			break;
569 		case RW_STATE_WORD0:
570 			count = pit_get_count(kvm, addr);
571 			ret = count & 0xff;
572 			s->read_state = RW_STATE_WORD1;
573 			break;
574 		case RW_STATE_WORD1:
575 			count = pit_get_count(kvm, addr);
576 			ret = (count >> 8) & 0xff;
577 			s->read_state = RW_STATE_WORD0;
578 			break;
579 		}
580 	}
581 
582 	if (len > sizeof(ret))
583 		len = sizeof(ret);
584 	memcpy(data, (char *)&ret, len);
585 
586 	mutex_unlock(&pit_state->lock);
587 	return 0;
588 }
589 
590 static int speaker_ioport_write(struct kvm_io_device *this,
591 				gpa_t addr, int len, const void *data)
592 {
593 	struct kvm_pit *pit = speaker_to_pit(this);
594 	struct kvm_kpit_state *pit_state = &pit->pit_state;
595 	struct kvm *kvm = pit->kvm;
596 	u32 val = *(u32 *) data;
597 	if (addr != KVM_SPEAKER_BASE_ADDRESS)
598 		return -EOPNOTSUPP;
599 
600 	mutex_lock(&pit_state->lock);
601 	pit_state->speaker_data_on = (val >> 1) & 1;
602 	pit_set_gate(kvm, 2, val & 1);
603 	mutex_unlock(&pit_state->lock);
604 	return 0;
605 }
606 
607 static int speaker_ioport_read(struct kvm_io_device *this,
608 			       gpa_t addr, int len, void *data)
609 {
610 	struct kvm_pit *pit = speaker_to_pit(this);
611 	struct kvm_kpit_state *pit_state = &pit->pit_state;
612 	struct kvm *kvm = pit->kvm;
613 	unsigned int refresh_clock;
614 	int ret;
615 	if (addr != KVM_SPEAKER_BASE_ADDRESS)
616 		return -EOPNOTSUPP;
617 
618 	/* Refresh clock toggles at about 15us. We approximate as 2^14ns. */
619 	refresh_clock = ((unsigned int)ktime_to_ns(ktime_get()) >> 14) & 1;
620 
621 	mutex_lock(&pit_state->lock);
622 	ret = ((pit_state->speaker_data_on << 1) | pit_get_gate(kvm, 2) |
623 		(pit_get_out(kvm, 2) << 5) | (refresh_clock << 4));
624 	if (len > sizeof(ret))
625 		len = sizeof(ret);
626 	memcpy(data, (char *)&ret, len);
627 	mutex_unlock(&pit_state->lock);
628 	return 0;
629 }
630 
631 void kvm_pit_reset(struct kvm_pit *pit)
632 {
633 	int i;
634 	struct kvm_kpit_channel_state *c;
635 
636 	mutex_lock(&pit->pit_state.lock);
637 	pit->pit_state.flags = 0;
638 	for (i = 0; i < 3; i++) {
639 		c = &pit->pit_state.channels[i];
640 		c->mode = 0xff;
641 		c->gate = (i != 2);
642 		pit_load_count(pit->kvm, i, 0);
643 	}
644 	mutex_unlock(&pit->pit_state.lock);
645 
646 	atomic_set(&pit->pit_state.pending, 0);
647 	pit->pit_state.irq_ack = 1;
648 }
649 
650 static void pit_mask_notifer(struct kvm_irq_mask_notifier *kimn, bool mask)
651 {
652 	struct kvm_pit *pit = container_of(kimn, struct kvm_pit, mask_notifier);
653 
654 	if (!mask) {
655 		atomic_set(&pit->pit_state.pending, 0);
656 		pit->pit_state.irq_ack = 1;
657 	}
658 }
659 
660 static const struct kvm_io_device_ops pit_dev_ops = {
661 	.read     = pit_ioport_read,
662 	.write    = pit_ioport_write,
663 };
664 
665 static const struct kvm_io_device_ops speaker_dev_ops = {
666 	.read     = speaker_ioport_read,
667 	.write    = speaker_ioport_write,
668 };
669 
670 /* Caller must hold slots_lock */
671 struct kvm_pit *kvm_create_pit(struct kvm *kvm, u32 flags)
672 {
673 	struct kvm_pit *pit;
674 	struct kvm_kpit_state *pit_state;
675 	struct pid *pid;
676 	pid_t pid_nr;
677 	int ret;
678 
679 	pit = kzalloc(sizeof(struct kvm_pit), GFP_KERNEL);
680 	if (!pit)
681 		return NULL;
682 
683 	pit->irq_source_id = kvm_request_irq_source_id(kvm);
684 	if (pit->irq_source_id < 0) {
685 		kfree(pit);
686 		return NULL;
687 	}
688 
689 	mutex_init(&pit->pit_state.lock);
690 	mutex_lock(&pit->pit_state.lock);
691 	spin_lock_init(&pit->pit_state.inject_lock);
692 
693 	pid = get_pid(task_tgid(current));
694 	pid_nr = pid_vnr(pid);
695 	put_pid(pid);
696 
697 	init_kthread_worker(&pit->worker);
698 	pit->worker_task = kthread_run(kthread_worker_fn, &pit->worker,
699 				       "kvm-pit/%d", pid_nr);
700 	if (IS_ERR(pit->worker_task)) {
701 		mutex_unlock(&pit->pit_state.lock);
702 		kvm_free_irq_source_id(kvm, pit->irq_source_id);
703 		kfree(pit);
704 		return NULL;
705 	}
706 	init_kthread_work(&pit->expired, pit_do_work);
707 
708 	kvm->arch.vpit = pit;
709 	pit->kvm = kvm;
710 
711 	pit_state = &pit->pit_state;
712 	pit_state->pit = pit;
713 	hrtimer_init(&pit_state->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
714 	pit_state->irq_ack_notifier.gsi = 0;
715 	pit_state->irq_ack_notifier.irq_acked = kvm_pit_ack_irq;
716 	kvm_register_irq_ack_notifier(kvm, &pit_state->irq_ack_notifier);
717 	pit_state->reinject = true;
718 	mutex_unlock(&pit->pit_state.lock);
719 
720 	kvm_pit_reset(pit);
721 
722 	pit->mask_notifier.func = pit_mask_notifer;
723 	kvm_register_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
724 
725 	kvm_iodevice_init(&pit->dev, &pit_dev_ops);
726 	ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, KVM_PIT_BASE_ADDRESS,
727 				      KVM_PIT_MEM_LENGTH, &pit->dev);
728 	if (ret < 0)
729 		goto fail;
730 
731 	if (flags & KVM_PIT_SPEAKER_DUMMY) {
732 		kvm_iodevice_init(&pit->speaker_dev, &speaker_dev_ops);
733 		ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS,
734 					      KVM_SPEAKER_BASE_ADDRESS, 4,
735 					      &pit->speaker_dev);
736 		if (ret < 0)
737 			goto fail_unregister;
738 	}
739 
740 	return pit;
741 
742 fail_unregister:
743 	kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->dev);
744 
745 fail:
746 	kvm_unregister_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
747 	kvm_unregister_irq_ack_notifier(kvm, &pit_state->irq_ack_notifier);
748 	kvm_free_irq_source_id(kvm, pit->irq_source_id);
749 	kthread_stop(pit->worker_task);
750 	kfree(pit);
751 	return NULL;
752 }
753 
754 void kvm_free_pit(struct kvm *kvm)
755 {
756 	struct hrtimer *timer;
757 
758 	if (kvm->arch.vpit) {
759 		kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &kvm->arch.vpit->dev);
760 		kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
761 					      &kvm->arch.vpit->speaker_dev);
762 		kvm_unregister_irq_mask_notifier(kvm, 0,
763 					       &kvm->arch.vpit->mask_notifier);
764 		kvm_unregister_irq_ack_notifier(kvm,
765 				&kvm->arch.vpit->pit_state.irq_ack_notifier);
766 		mutex_lock(&kvm->arch.vpit->pit_state.lock);
767 		timer = &kvm->arch.vpit->pit_state.timer;
768 		hrtimer_cancel(timer);
769 		flush_kthread_work(&kvm->arch.vpit->expired);
770 		kthread_stop(kvm->arch.vpit->worker_task);
771 		kvm_free_irq_source_id(kvm, kvm->arch.vpit->irq_source_id);
772 		mutex_unlock(&kvm->arch.vpit->pit_state.lock);
773 		kfree(kvm->arch.vpit);
774 	}
775 }
776